Why CIMILab stands out
Our work is shaped by open-science habits, strong evaluation, and a clear focus on research that can matter beyond the benchmark.
Versioned data, documented training settings, and scripted evaluation make our experiments inspectable and repeatable.
We connect foundational ML methods to medical imaging, biomedical NLP, and practical deployment questions.
Our projects and publications are structured to invite collaborators, students, and external partners into the work.
Research Focus
Deep learning architectures, ensemble methods, and attention mechanisms for high-stakes prediction and segmentation tasks.
Data infrastructure, benchmarking frameworks, and evaluation pipelines that make AI experiments reproducible and comparable.
Uncertainty quantification, interpretability methods, and fairness auditing — building AI that clinicians and patients can trust.
Bridging research and deployment — efficient inference, edge adaptation, and clinical decision support at real-world scale.
Flagship Research
Interpretable multi-architecture ensemble for ischemic stroke lesion segmentation on the ISLES benchmark.
Dual encoder representation network for spinal lesion detection and classification on VinDr-SpineXR.
Biomedical foundation model fine-tuning and evaluation suite for clinical NLP tasks including QA, summarization, and NER.
Lab Updates
CIMILab is proud to announce the official launch of the BioMed-LLM repository — a suite of fine-tuned large language models for biomedical NLP tasks.
Our work on biomedical LLM evaluation has been accepted to the BioNLP Workshop at ACL 2025.
CIMILab is thrilled to welcome two new graduate researchers joining for Spring 2025: Arpita Howlader and Samrat Kumar Dey.